Researcher(s)
- Vishv Kansagara, Finance, University of Delaware
Faculty Mentor(s)
- Gerard Pinto, Department of Finance, University of Delaware
Abstract
Between 2023 and 2026 a small number of AI infrastructure listings, among them Arm, CoreWeave and Cerebras, absorbed an unusually large share of the capital raised in US initial public offerings. This study asks whether that concentration came at the expense of everyone else: whether companies outside technology that went public in the same windows faced a thinner calendar and weaker investor demand. The question is one of relative crowding within the IPO market rather than about the total size of that market, a distinction the available data can identify.
The sample is 1737 IPOs priced on US exchanges between 2015 and 2026, drawn from the SDC New Issues database and screened to mirror Ritter’s published sample filters, with offer prices matched to first closing prices in Compustat daily security data. AI infrastructure issuers come from a list of firms fixed and verified against their SEC prospectuses before any estimation, because an industry code screen was shown to capture ordinary semiconductor listings while missing the AI cloud firms. Quarterly AI infrastructure proceeds are scaled by aggregate US equity market value at the start of the quarter. That denominator sits outside the IPO market by construction, so unlike a share of contemporaneous issuance it cannot mechanically manufacture the crowding it is meant to detect. Two tests follow: whether quarterly issuance by companies outside technology falls as AI intensity rises, and whether such firms pricing in those quarters show smaller first day returns, the standard proxy for demand revealed during bookbuilding. Because AI infrastructure proceeds are structurally zero before late 2023 and concentrated in a handful of quarters afterward, estimates are reported as descriptive rather than causal, under both a narrow and a broad definition of AI infrastructure.